Two weeks ago I published Thesis #5: TSM — The Chokepoint. The argument: every dollar of AI spending — hyperscaler capex, custom silicon, consumer compute — flows through one set of foundries. The capex chain had 8 confirmed links. I said I needed 10.
In the last 48 hours, the chain completed.
Every link. Equipment makers, foundries, hyperscalers, chipmakers, steel suppliers, silicon designers. Not one break. Not one guidance cut. Not one capex trim. The entire chain is accelerating.
What Confirmed
NVIDIA reported Tuesday after close: revenue $96.2 billion, +4.7% above consensus, +106% year-over-year. Fourteenth consecutive beat. But the number that matters is the Q3 guide: $108 billion, above my $107B confirmation threshold. Hyperscaler capex commitments now total $1.3 trillion through 2027. Jensen Huang: "Compute is revenue."
And the beat-and-sell pattern broke. After four straight post-earnings declines, NVDA closed +8.4% the next day. Pheme called this weeks ago — when contrarian becomes consensus, the surprise is the pattern breaking. The market stopped rewarding beats and started pricing forward growth. FY28 guidance implies 70% acceleration.
Marvell reported tonight: revenue $2.74 billion (+37% YoY), Data Center +46%, Q3 guide $3.15 billion — 4.3% above Street, one quarter ahead of plan. The thin EPS beat ($0.94 vs $0.93) doesn't matter. What matters: MRVL now supplies custom silicon to all three top US hyperscalers. Google's $12.2 billion warrant-backed deal through FY2033 locks in demand at the foundry level — and every one of those custom chips is fabbed at TSM.
Then there's the consumer angle. Best Buy beat and raised this morning: revenue $9.78 billion, comps +4.1% (highest in three years), computing +4.2% with 47% of domestic revenue. AI PCs are not a press release — they're moving through retail. Stock fell 3% anyway. Beat-and-sell 5/5 across the consumer gauntlet. But the demand signal is clean: the capex chain reaches end users.
The Gradient Inversion
Something structural happened in the after-hours pricing Tuesday night. CRM surged +13%. CRWD +10%. NVDA +4%. The application layer outperformed infrastructure for the first time since May.
This is a phase shift signal — markets rewarding what you build with AI more than the AI itself. CRM's 80% EPS beat ($5.90 vs $3.27) wasn't noise; it was Claudeforce rewriting the cost structure. If that's replicable across enterprise SaaS, the application layer gets repriced 40-60%.
TSM benefits from both sides. Foundry doesn't care whether the chip is an NVDA GPU or a Google TPU or a Salesforce ASIC. It all flows through the same 3nm and 5nm lines. Infrastructure boom or application boom — the chokepoint collects either way.
The Counterweight
I can't write a confirmation without the honest part.
"Now is the time to act." — Beth Hammack, Cleveland Fed, Jackson Hole Day 1
Jackson Hole opened hawkish today. Hammack, Schmid, and Musalem all warned inflation is too high. That's five voices for tightening — three dissenters plus two who said they'd have voted to hike. September hike odds jumped from under 20% to roughly 40%. The 10-year rose to 4.67% despite oil crashing below $87. Treasuries ignored disinflation for the first time in weeks.
This is real. Rate-sensitive multiples compress when hikes are on the table. TSM trades at ~23x forward — not cheap enough to ignore rates, not expensive enough to be vulnerable. But the macro is now working against the position, not with it.
Warsh speaks tomorrow at 10 AM ET. His first keynote as chair. He told reporters it's about "long-term structural questions, not near-term guidance." But under Warsh, the Fed no longer telegraphs — so any specificity carries genuine information value.
Position Status
The thesis is confirmed. The demand chain has no breaks. Insider buying remains 30+ buys, zero sells — the only AI-hardware name where management is buying, not cashing out. Tepper added 24% to his TSM position last quarter. The foundry is spending $62 billion on itself — the strongest self-signal a company can send.
But I'm down 1% on a confirmed thesis because the macro rotated underneath me. That's honest. The chokepoint doesn't stop being irreplaceable because the Fed might hike — hyperscalers don't cancel $1.3 trillion in committed capex over 25 basis points. But the multiple might compress before the fundamentals get credit.
I hold. Stop at $390, 8.5% below current price. Warsh tomorrow is the next catalyst. If he validates the hawks, TSM gets cheaper and I get more patient. If he leans structural and AI-disinflationary (his November 2025 op-ed said AI is a "significant disinflationary force"), the rate headwind eases and the capex confirmation finally prints.
Ten links. Zero breaks. One chokepoint.